1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MEMREMAP_H_
3 #define _LINUX_MEMREMAP_H_
4
5 #include <linux/mm.h>
6 #include <linux/range.h>
7 #include <linux/ioport.h>
8 #include <linux/percpu-refcount.h>
9
10 struct resource;
11 struct device;
12
13 /**
14 * struct vmem_altmap - pre-allocated storage for vmemmap_populate
15 * @base_pfn: base of the entire dev_pagemap mapping
16 * @reserve: pages mapped, but reserved for driver use (relative to @base)
17 * @free: free pages set aside in the mapping for memmap storage
18 * @align: pages reserved to meet allocation alignments
19 * @alloc: track pages consumed, private to vmemmap_populate()
20 */
21 struct vmem_altmap {
22 unsigned long base_pfn;
23 const unsigned long end_pfn;
24 const unsigned long reserve;
25 unsigned long free;
26 unsigned long align;
27 unsigned long alloc;
28 };
29
30 /*
31 * Specialize ZONE_DEVICE memory into multiple types each has a different
32 * usage.
33 *
34 * MEMORY_DEVICE_PRIVATE:
35 * Device memory that is not directly addressable by the CPU: CPU can neither
36 * read nor write private memory. In this case, we do still have struct pages
37 * backing the device memory. Doing so simplifies the implementation, but it is
38 * important to remember that there are certain points at which the struct page
39 * must be treated as an opaque object, rather than a "normal" struct page.
40 *
41 * A more complete discussion of unaddressable memory may be found in
42 * include/linux/hmm.h and Documentation/vm/hmm.rst.
43 *
44 * MEMORY_DEVICE_FS_DAX:
45 * Host memory that has similar access semantics as System RAM i.e. DMA
46 * coherent and supports page pinning. In support of coordinating page
47 * pinning vs other operations MEMORY_DEVICE_FS_DAX arranges for a
48 * wakeup event whenever a page is unpinned and becomes idle. This
49 * wakeup is used to coordinate physical address space management (ex:
50 * fs truncate/hole punch) vs pinned pages (ex: device dma).
51 *
52 * MEMORY_DEVICE_GENERIC:
53 * Host memory that has similar access semantics as System RAM i.e. DMA
54 * coherent and supports page pinning. This is for example used by DAX devices
55 * that expose memory using a character device.
56 *
57 * MEMORY_DEVICE_PCI_P2PDMA:
58 * Device memory residing in a PCI BAR intended for use with Peer-to-Peer
59 * transactions.
60 */
61 enum memory_type {
62 /* 0 is reserved to catch uninitialized type fields */
63 MEMORY_DEVICE_PRIVATE = 1,
64 MEMORY_DEVICE_FS_DAX,
65 MEMORY_DEVICE_GENERIC,
66 MEMORY_DEVICE_PCI_P2PDMA,
67 };
68
69 struct dev_pagemap_ops {
70 /*
71 * Called once the page refcount reaches 0. The reference count will be
72 * reset to one by the core code after the method is called to prepare
73 * for handing out the page again.
74 */
75 void (*page_free)(struct page *page);
76
77 /*
78 * Used for private (un-addressable) device memory only. Must migrate
79 * the page back to a CPU accessible page.
80 */
81 vm_fault_t (*migrate_to_ram)(struct vm_fault *vmf);
82 };
83
84 #define PGMAP_ALTMAP_VALID (1 << 0)
85
86 /**
87 * struct dev_pagemap - metadata for ZONE_DEVICE mappings
88 * @altmap: pre-allocated/reserved memory for vmemmap allocations
89 * @ref: reference count that pins the devm_memremap_pages() mapping
90 * @done: completion for @ref
91 * @type: memory type: see MEMORY_* in memory_hotplug.h
92 * @flags: PGMAP_* flags to specify defailed behavior
93 * @vmemmap_shift: structural definition of how the vmemmap page metadata
94 * is populated, specifically the metadata page order.
95 * A zero value (default) uses base pages as the vmemmap metadata
96 * representation. A bigger value will set up compound struct pages
97 * of the requested order value.
98 * @ops: method table
99 * @owner: an opaque pointer identifying the entity that manages this
100 * instance. Used by various helpers to make sure that no
101 * foreign ZONE_DEVICE memory is accessed.
102 * @nr_range: number of ranges to be mapped
103 * @range: range to be mapped when nr_range == 1
104 * @ranges: array of ranges to be mapped when nr_range > 1
105 */
106 struct dev_pagemap {
107 struct vmem_altmap altmap;
108 struct percpu_ref ref;
109 struct completion done;
110 enum memory_type type;
111 unsigned int flags;
112 unsigned long vmemmap_shift;
113 const struct dev_pagemap_ops *ops;
114 void *owner;
115 int nr_range;
116 union {
117 struct range range;
118 struct range ranges[0];
119 };
120 };
121
pgmap_altmap(struct dev_pagemap * pgmap)122 static inline struct vmem_altmap *pgmap_altmap(struct dev_pagemap *pgmap)
123 {
124 if (pgmap->flags & PGMAP_ALTMAP_VALID)
125 return &pgmap->altmap;
126 return NULL;
127 }
128
pgmap_vmemmap_nr(struct dev_pagemap * pgmap)129 static inline unsigned long pgmap_vmemmap_nr(struct dev_pagemap *pgmap)
130 {
131 return 1 << pgmap->vmemmap_shift;
132 }
133
is_device_private_page(const struct page * page)134 static inline bool is_device_private_page(const struct page *page)
135 {
136 return IS_ENABLED(CONFIG_DEVICE_PRIVATE) &&
137 is_zone_device_page(page) &&
138 page->pgmap->type == MEMORY_DEVICE_PRIVATE;
139 }
140
folio_is_device_private(const struct folio * folio)141 static inline bool folio_is_device_private(const struct folio *folio)
142 {
143 return is_device_private_page(&folio->page);
144 }
145
is_pci_p2pdma_page(const struct page * page)146 static inline bool is_pci_p2pdma_page(const struct page *page)
147 {
148 return IS_ENABLED(CONFIG_PCI_P2PDMA) &&
149 is_zone_device_page(page) &&
150 page->pgmap->type == MEMORY_DEVICE_PCI_P2PDMA;
151 }
152
153 #ifdef CONFIG_ZONE_DEVICE
154 void *memremap_pages(struct dev_pagemap *pgmap, int nid);
155 void memunmap_pages(struct dev_pagemap *pgmap);
156 void *devm_memremap_pages(struct device *dev, struct dev_pagemap *pgmap);
157 void devm_memunmap_pages(struct device *dev, struct dev_pagemap *pgmap);
158 struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
159 struct dev_pagemap *pgmap);
160 bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn);
161
162 unsigned long vmem_altmap_offset(struct vmem_altmap *altmap);
163 void vmem_altmap_free(struct vmem_altmap *altmap, unsigned long nr_pfns);
164 unsigned long memremap_compat_align(void);
165 #else
devm_memremap_pages(struct device * dev,struct dev_pagemap * pgmap)166 static inline void *devm_memremap_pages(struct device *dev,
167 struct dev_pagemap *pgmap)
168 {
169 /*
170 * Fail attempts to call devm_memremap_pages() without
171 * ZONE_DEVICE support enabled, this requires callers to fall
172 * back to plain devm_memremap() based on config
173 */
174 WARN_ON_ONCE(1);
175 return ERR_PTR(-ENXIO);
176 }
177
devm_memunmap_pages(struct device * dev,struct dev_pagemap * pgmap)178 static inline void devm_memunmap_pages(struct device *dev,
179 struct dev_pagemap *pgmap)
180 {
181 }
182
get_dev_pagemap(unsigned long pfn,struct dev_pagemap * pgmap)183 static inline struct dev_pagemap *get_dev_pagemap(unsigned long pfn,
184 struct dev_pagemap *pgmap)
185 {
186 return NULL;
187 }
188
pgmap_pfn_valid(struct dev_pagemap * pgmap,unsigned long pfn)189 static inline bool pgmap_pfn_valid(struct dev_pagemap *pgmap, unsigned long pfn)
190 {
191 return false;
192 }
193
vmem_altmap_offset(struct vmem_altmap * altmap)194 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
195 {
196 return 0;
197 }
198
vmem_altmap_free(struct vmem_altmap * altmap,unsigned long nr_pfns)199 static inline void vmem_altmap_free(struct vmem_altmap *altmap,
200 unsigned long nr_pfns)
201 {
202 }
203
204 /* when memremap_pages() is disabled all archs can remap a single page */
memremap_compat_align(void)205 static inline unsigned long memremap_compat_align(void)
206 {
207 return PAGE_SIZE;
208 }
209 #endif /* CONFIG_ZONE_DEVICE */
210
put_dev_pagemap(struct dev_pagemap * pgmap)211 static inline void put_dev_pagemap(struct dev_pagemap *pgmap)
212 {
213 if (pgmap)
214 percpu_ref_put(&pgmap->ref);
215 }
216
217 #endif /* _LINUX_MEMREMAP_H_ */
218